摘要：

海上风电作为可再生能源的重要来源, 在近年来得到了广泛关注和迅速发展。海上风电的强随机性和波动性对传统预测方法的准确性提出了挑战。考虑到海上风力的季节性规律, 从信号分析的角度出发, 设计一种全新频域约束下的海上风电超短期功率深度预测模型。引入一个极简的框架, 结合时间序列分解、频率加权和线性模型, 并提出一种频域驱动的季节性分解方法, 构建核函数和季节性因子来对多尺度季节性进行建模。在此基础上提出一种频率加权机制来动态增强或抑制季节性频率分量。除更高效外, 该模型的空间复杂度与序列成线性比例, 并产生最少数量的参数。基于Alpha Ventus海上风场实际运行数据进行相关实验, 结果表明, 通过与最先进的预测模型进行比较, 证明了模型的有效性和出色的性能。另外进行消融实验及可视化以促进对模型的理解。

关键词: 海上风电, 功率预测, 季节性时间序列, 信号分析, 时间序列分解

Abstract:

Offshore wind power, a significant source of renewable energy, has garnered widespread attention and experienced rapid development in recent years. The strong stochasticity and volatility of offshore wind power hinder the accuracy of traditional prediction methods. To address this challenge, this study considers the seasonal patterns of wind power and proposes a new ultra-short-term power depth prediction model for offshore wind power under frequency domain constraints, from the perspective of signal analysis. First, the study introduces an ultrasimple framework that combines time series decomposition, frequency weighting, and linear modeling. Second, the study proposes a frequency domain driven seasonal decomposition method to construct kernel functions and seasonal factors, to model multiscale seasonality. Third, the study presents a frequency weighting mechanism to dynamically enhance or suppress the seasonal frequency components. In addition to being more efficient, the space complexity of the proposed model is linearly proportional to the sequence, and it generates the smallest number of parameters. Finally, experiments on real operational data from the Alpha Ventus offshore wind farm confirm the effectiveness and outstanding performance of the proposed model compared to those of state-of-the-art prediction models. Ablation studies and visualizations are provided to facilitate the understanding of the proposed approach.

Key words: offshore wind power, power forecasting, seasonal time series, signal analysis, time series decomposition